Computer Science

Risk and response communication of public agencies through social media played a significant role in the emergence and spread of novel Coronavirus (COVID-19) and such interactions were echoed in other information outlets. This study collected time-sensitive online social media data and analyzed such communication patterns from public health (WHO, CDC), emergency (FEMA), and transportation (FDOT) agencies using data-driven methods. The scope of the work includes a detailed understanding of how agencies communicate risk information through social media during a pandemic and influence community response (i.e. timing of lockdown, timing of reopening) and disease outbreak indicators (i.e. number of confirmed cases, number of deaths). The data includes Twitter interactions from different agencies (2.15K tweets per agency on average) and crowdsourced data (i.e. Worldometer) on COVID-19 cases and deaths were observed between February 21, 2020 and June 06, 2020. Several machine learning techniques such as (i.e. topic mining and sentiment ratings over time) are applied here to identify the dynamics of emergent topics during this unprecedented time. Temporal infographics of the results captured the agency-levels variations over time in circulating information about the importance of face covering, home quarantine, social distancing and contact tracing. In addition, agencies showed differences in their discussions about community transmission, lack of personal protective equipment, testing and medical supplies, use of tobacco, vaccine, mental health issues, hospitalization, hurricane season, airports, construction work among others. Findings could support more efficient transfer of risk and response information as communities shift to new normal as well as in future pandemics.

The intrinsic temporality of learning demands the adoption of methodologies capable of exploiting time-series information. In this study we leverage the sequence data framework and show how data-driven analysis of temporal sequences of task completion in online courses can be used to characterise personal and group learners' behaviors, and to identify critical tasks and course sessions in a given course design. We also introduce a recently developed probabilistic Bayesian model to learn sequence trajectories of students and predict student performance. The application of our data-driven sequence-based analyses to data from learners undertaking an on-line Business Management course reveals distinct behaviors within the cohort of learners, identifying learners or groups of learners that deviate from the nominal order expected in the course. Using course grades a posteriori, we explore differences in behavior between high and low performing learners. We find that high performing learners follow the progression between weekly sessions more regularly than low performing learners, yet within each weekly session high performing learners are less tied to the nominal task order. We then model the sequences of high and low performance students using the probablistic Bayesian model and show that we can learn engagement behaviors associated with performance. We also show that the data sequence framework can be used for task centric analysis; we identify critical junctures and differences among types of tasks within the course design. We find that non-rote learning tasks, such as interactive tasks or discussion posts, are correlated with higher performance. We discuss the application of such analytical techniques as an aid to course design, intervention, and student supervision.

There are some shreds of evidence that social opinion polarization leads to the breakup of the relationship, some in the scale of small communities, but others can divide large organizations or even a nation. The legacy methodology to answer the root cause of opinion polarization in the society commonly use random sampling or questionnaire-based approach, which considered expensive in term of time and money. On the contrary, the opportunity to employ big data approach using social media data provides a rich source to investigate several questions, such as: how social opinion polarization formed, dynamic social network mechanisms during time-windows observation, identification of dominant actors and communities. The power of the big data approach lies in the number of data analyzed, where the more data involved in the process, the more accurate to describe the population condition. Today, computing power is no longer become an obstacle to process large-scale data, thus the observation of the social opinion polarization process is possible.

Node representation learning for signed directed networks has received considerable attention in many real-world applications such as link sign prediction, node classification and node recommendation. The challenge lies in how to adequately encode the complex topological information of the networks. Recent studies mainly focus on preserving the first-order network topology which indicates the closeness relationships of nodes. However, these methods generally fail to capture the high-order topology which indicates the local structures of nodes and serves as an essential characteristic of the network topology. In addition, for the first-order topology, the additional value of non-existent links is largely ignored. In this paper, we propose to learn more representative node embeddings by simultaneously capturing the first-order and high-order topology in signed directed networks. In particular, we reformulate the representation learning problem on signed directed networks from a variational auto-encoding perspective and further develop a decoupled variational embedding (DVE) method. DVE leverages a specially designed auto-encoder structure to capture both the first-order and high-order topology of signed directed networks, and thus learns more representative node embedding. Extensive experiments are conducted on three widely used real-world datasets. Comprehensive results on both link sign prediction and node recommendation task demonstrate the effectiveness of DVE. Qualitative results and analysis are also given to provide a better understanding of DVE.

Veracity of data posted on the microblog platforms has in recent years been a subject of intensive study by professionals specializing in various fields of informatics as well as sociology, particularly in the light of increasing importance of online tools for news spreading. On Twitter and similar sites, it is possible to report on ongoing situations globally with minimal delay, while the cost of such reporting remains negligible. One of the most important features of this social network is that content delivery can be customized to allow users to focus only on news items covering subject matters they find interesting. With this in mind, it becomes necessary to create verification mechanisms that can ascertain whether the claims made on Twitter can be taken seriously and prevent false content from spreading too far. This study demonstrates an innovative System for verification of information that can fulfill the role described above. The System is comprised of four mutually connected modules: a legacy module, a trustworthiness classifier; a module managing user authority, and a ranking procedure. All of the modules function within an integrated framework and jointly contribute to an accurate classification of messages and authors. Effectiveness of the solution was evaluated empirically on a sample of Twitter users, with a strict 10-fold evaluation procedure applied for each module. The findings indicate that the solution successfully meets the primary objectives of the study and performs its function as expected.

Alarmed by the volume of disinformation that was assumed to have taken place during the 2016 US elections, scholars, politics and journalists predicted the worst when the first deepfakes began to emerge in 2018. After all, US Elections 2020 were believed to be the most secure in American history. This paper seeks explanations for an apparent contradiction: we believe that it was precisely the multiplication and conjugation of different types of warnings and fears that created the conditions that prevented malicious political deepfakes from affecting the 2020 US elections. From these warnings, we identified four factors (more active role of social networks, new laws, difficulties in accessing Artificial Intelligence and better awareness of society). But while this formula has proven to be effective in the case of the United States, 2020, it is not correct to assume that it can be repeated in other political contexts.

Despite the growing interest in characterizing the local geometry leading to the global topology of networks, our understanding of the local structure of complex networks, especially real-world networks, is still incomplete. Here, we analyze a simple, elegant yet underexplored measure, `degree difference' (DD) between vertices of an edge, to understand the local network geometry. We describe the connection between DD and global assortativity of the network from both formal and conceptual perspective, and show that DD can reveal structural properties that are not obtained from other such measures in network science. Typically, edges with different DD play different structural roles and the DD distribution is an important network signature. Notably, DD is the basic unit of assortativity. We provide an explanation as to why DD can characterize structural heterogeneity in mixing patterns unlike global assortativity and local node assortativity. By analyzing synthetic and real networks, we show that DD distribution can be used to distinguish between different types of networks including those networks that cannot be easily distinguished using degree sequence and global assortativity. Moreover, we show DD to be an indicator for topological robustness of scale-free networks. Overall, DD is a local measure that is simple to define, easy to evaluate, and that reveals structural properties of networks not readily seen from other measures.

The networked opinion diffusion in online social networks (OSN) is often governed by the two genres of opinions - endogenous opinions that are driven by the influence of social contacts among users, and exogenous opinions which are formed by external effects like news, feeds etc. Accurate demarcation of endogenous and exogenous messages offers an important cue to opinion modeling, thereby enhancing its predictive performance. In this paper, we design a suite of unsupervised classification methods based on experimental design approaches, in which, we aim to select the subsets of events which minimize different measures of mean estimation error. In more detail, we first show that these subset selection tasks are NP-Hard. Then we show that the associated objective functions are weakly submodular, which allows us to cast efficient approximation algorithms with guarantees. Finally, we validate the efficacy of our proposal on various real-world datasets crawled from Twitter as well as diverse synthetic datasets. Our experiments range from validating prediction performance on unsanitized and sanitized events to checking the effect of selecting optimal subsets of various sizes. Through various experiments, we have found that our method offers a significant improvement in accuracy in terms of opinion forecasting, against several competitors.

The idea underlying the modal formulation of density-based clustering is to associate groups with the regions around the modes of the probability density function underlying the data. This correspondence between clusters and dense regions in the sample space is here exploited to discuss an extension of this approach to the analysis of social networks. Such extension seems particularly appealing: conceptually, the notion of high-density cluster fits well the one of community in a network, regarded to as a collection of individuals with dense local ties in its neighbourhood. The lack of a probabilistic notion of density in networks is turned into a major strength of the proposed method, where node-wise measures that quantify the role and position of actors may be used to derive different community configurations. The approach allows for the identification of a hierarchical structure of clusters, which may catch different degrees of resolution of the clustering structure. This feature well fits the nature of social networks, disentangling a different involvement of individuals in social aggregations.

The COVID-19 pandemic is having a serious adverse impact on the lives of people across the world. COVID-19 has exacerbated community-wide depression, and has led to increased drug abuse brought about by isolation of individuals as a result of lockdown. Further, apart from providing informative content to the public, the incessant media coverage of COVID-19 crisis in terms of news broadcasts, published articles and sharing of information on social media have had the undesired snowballing effect on stress levels (further elevating depression and drug use) due to uncertain future. In this position paper, we propose a novel framework for assessing the spatio-temporal-thematic progression of depression, drug abuse, and informativeness of the underlying news content across the different states in the United States. Our framework employs an attention-based transfer learning technique to apply knowledge learned on a social media domain to a target domain of media exposure. To extract news articles that are related to COVID-19 communications from the streaming news content on the web, we use neural semantic parsing, and background knowledge bases in a sequence of steps called semantic filtering. We achieve promising preliminary results on three variations of Bidirectional Encoder Representations from Transformers (BERT) model. We compare our findings against a report from Mental Health America and the results show that our fine-tuned BERT models perform better than vanilla BERT. Our study can benefit epidemiologists by offering actionable insights on COVID-19 and its regional impact. Further, our solution can be integrated into end-user applications to tailor news for users based on their emotional tone measured on the scale of depressiveness, drug abusiveness, and informativeness.